feat: test the oracle against its own records

Every reference block's run, judged again with the oracle as committed, boots in each repetition. Beetle Saturn's three BIOS lines are BIOS errors, the Japanese menu screens are error screens, and the PlayStation cores carry no error pattern.

tests/test_core_probe.py

@@ -286,6 +286,45 @@ class OracleTests(unittest.TestCase):
         self.assertEqual(core_probe.oracle_for(data, "b" * 40, "Fake")["references"], [])
         self.assertEqual(core_probe.oracle_for(data, "a" * 40, "Nobody")["error_patterns"], ["shared"])
 
+    def test_every_reference_judges_its_own_record_a_boot(self) -> None:
+        """Each reference block names the run it was taken from: judged
+        again with the oracle as it stands, every repetition of that run
+        boots, so a frame copied wrong or a stage mislabelled fails here."""
+        data, _ = core_probe.load_oracle(core_probe.DEFAULT_ORACLE)
+        records = {s.record["id"]: s.record for s in measurement_records.read_records(REPO_ROOT)}
+        checked = 0
+        for item in data.get("content") or ():
+            for library, block in item["references"].items():
+                record = records.get(block["recorded_from"])
+                self.assertIsNotNone(record, f"{item['name']} on {library}: {block['recorded_from']}")
+                self.assertEqual(record["inputs"]["content"]["sha1"], item["sha1"], item["name"])
+                for rep in record["variants"][0]["repetitions"]:
+                    oracle = core_probe.oracle_for(data, item["sha1"], rep["library_name"])
+                    verdict, evidence = core_probe.judge(rep, oracle)
+                    self.assertEqual(verdict, "boots", f"{item['name']} on {library}: {evidence}")
+                    checked += 1
+        self.assertGreaterEqual(checked, 2 * 14)
+
+    def test_the_repository_error_patterns_and_screens(self) -> None:
+        data, _ = core_probe.load_oracle(core_probe.DEFAULT_ORACLE)
+        saturn = core_probe.oracle_for(data, None, "Beetle Saturn")
+        for line in ('Cannot open BIOS file "/system/sega_101.bin".',
+                     'BIOS file "/system/sega_101.bin" is of an incorrect size.',
+                     'BIOS file "/system/sega_101.bin" could not be fully read (short or failed read).'):
+            rep = {"loaded": False, "logs": [{"frame": 0, "level": "error", "text": line}]}
+            self.assertEqual(core_probe.judge(rep, saturn)[0], "bios_error", line)
+        menu = {"Beetle PSX": "dfbfc299250c0983f86bbf951efd187441306d92",
+                "Beetle PSX HW": "dfbfc299250c0983f86bbf951efd187441306d92",
+                "SwanStation": "167fb7b6527e295b12dca5806a29856f5f31d401"}
+        for library, sha1 in menu.items():
+            rep = {"loaded": True, "logs": [], "frames": 100, "warmup": 0,
+                   "observables": {"frame_hashes": [[90, sha1]], "uniform_frames": []}}
+            verdict, evidence = core_probe.judge(rep, core_probe.oracle_for(data, None, library))
+            self.assertEqual((verdict, evidence["rule"]), ("bios_error", "error_screen"), library)
+        # The PlayStation cores fall back to a built-in BIOS: nothing they log is an error.
+        for library in ("Beetle PSX", "SwanStation", "PCSX-ReARMed", "PCSX1"):
+            self.assertEqual(core_probe.oracle_for(data, None, library)["error_patterns"], [])
+
     def test_the_repository_oracle_is_well_formed(self) -> None:
         import yaml
         data = yaml.safe_load(core_probe.DEFAULT_ORACLE.read_text(encoding="utf-8"))